Apple is still selling the A7-powered iPad Air at the same price as the iPad Mini 3 [Retina], so they probably didn't want to create a scenario where people would choose one over the other based on application processor.
Apple A8 and A8X
- Thread starter ltcommander.data
- Start date
Apple is still selling the A7-powered iPad Air at the same price as the iPad Mini 3 [Retina], so they probably didn't want to create a scenario where people would choose one over the other based on application processor.
Such thinness has the potential to feel really comfortable and modern for a tablet profile, though its ability to dampen the feel after each tap when typing and navigating will probably lessen even more. Not so much a negative, though, as it is a new tactile characteristic of a more modern device profile.
The internals, like the display and SoC performance, also have to be quite efficient to achieve their claims within such a narrow space. The approach of nVidia's architecture, relying on comparatively high clocks, and presumably voltages, for their Denver CPU cores and other processors, may not leave them as much headroom.
The internals, like the display and SoC performance, also have to be quite efficient to achieve their claims within such a narrow space. The approach of nVidia's architecture, relying on comparatively high clocks, and presumably voltages, for their Denver CPU cores and other processors, may not leave them as much headroom.
Well I think you've got it wrong regarding the CPU core. Cyclone is already a pretty power hungry CPU core (relatively speaking, compared to most other mobile CPU cores), while Denver eschews the power hungry OoO logic and tries to achieve high performance and efficiency in a very different way by optimizing once, reusing many times, and executing optimized code in-order. Anyway, this would be an apples to oranges comparison of power consumption because A8 and A8X are using a more advanced 20nm fab. process node while NVIDIA will be moving to that node in the first half of next year. And FWIW, higher clocks does not automatically equate to lower efficiency (see Maxwell vs. Kepler GPU and Denver vs. Cortex A15 CPU as some examples of that).
Last edited by a moderator:
DuckThor Evil
Legend
I don't think there is much crossover between those two, very different products.
The chassis's ability to dissipate heat is important too when it comes to performance degradation. For instance, Shield tablet has virtually no degradation of either CPU or GPU performance after 100 continuously looped runs due to the internal magnesium shield (most tablets, including iPad's, have significant throttling of CPU and/or GPU clock operating frequencies over time). We will have to see how well the Nexus 9 does in this area (and mobile Kepler obviously cannot rely on lower precision FP16 shaders too because it always renders at the highest quality).
anexanhume
Veteran
Historically iPad CPU speeds have been 100 to 200 MHz more than comparable iPhone CPU speeds.
The battery capacity in iPad Air is not small by any means (for instance, it has ~ 32% more battery capacity than the Nexus 9). That said, the battery capacity in iPad Mini Retina is much smaller in comparison, while achieving similar battery life as iPad Air. So the screen size and overall platform power consumption really makes a big difference.
On 28nm, Cyclone had very good performance and very competitive perf. per watt compared to something like R3 Cortex A15, so Cyclone enhanced on 20nm should be even better no doubt. Who really knows how it compares to Denver though.
X2.5 iPad air.
A8 was up to 50% extra compare to A7
So I make that A8X is x166% of A8. So 50% from extra 2 clusters, and some from extra clock frequency.
Beyond the GPU I wonder where the extra 1B transistors are going, one assumes a double width memory controller?
I assume there was no mention of ram. Also, strange that there was a lot of rumours of increased pixel density, but that never materialised.
Regarding powerVR series 7, I think it's safe to reckon that although it is yet to be announced, lead partners (ie Apple) already have it.
A8 was up to 50% extra compare to A7
So I make that A8X is x166% of A8. So 50% from extra 2 clusters, and some from extra clock frequency.
Beyond the GPU I wonder where the extra 1B transistors are going, one assumes a double width memory controller?
I assume there was no mention of ram. Also, strange that there was a lot of rumours of increased pixel density, but that never materialised.
Regarding powerVR series 7, I think it's safe to reckon that although it is yet to be announced, lead partners (ie Apple) already have it.
anexanhume
Veteran
Leaked logic board shows 2GB RAM modules. http://www.macrumors.com/2014/10/13/ipad-air-2-a8x-2-gb-ram/
I am curious if they returned to 128 bit too. I assume they need to to keep that GPU fed.
Do you really think GPU clock operating frequency increased too? Going from A7 to A8, GPU performance in T-Rex HD Offscreen test increased by ~ 1.62x. So assuming that GPU performance increases by ~1.5x going from A8 to A8X, that would be ~ 2.43x improvement for A8X vs. A7. You could be right, but it doesn't seem that GPU clock operating frequency will need to increase much here.
Last edited by a moderator:
Well I guess its all estimates and judgement calls, and frankly just something to chat about.
I know for the A8 they said "up to 50%".
I didn't see the presentation this evening, just the slide that said "x2.5". I don't know if that was "up to" as well in the narrative.
How they are defining that x2.5 is of course unknown, It's possible the frequency could be the same. Given that they are known to tinker with the GPU frequency (iphone 5s/ipad air and iphone 6/6+), then it is certainly in their MO to do a slight frequency adjustment.
My best guess has always been that the iPad Air's A7 was a 467 MHz G6430 + 1.4 GHz Cyclone and that a 500 MHz GX6650 + 1.5 GHz Cyclone v2 as I speculated the other day would be a logical extrapolation for an A8X (once evidence of its existence appeared... before that, I just assumed they'd roll with the A8 for the new iPads.)
I still think it could be. Depending upon whether the graphics benchmark results skew marginally higher than I'm expecting, a higher clock of maybe 533 MHz (probably not higher than 567 MHz, though) might fit for the GPU with a 1.6 GHz clock (with an outside chance of 1.7 GHz) for the CPU.
I still think it could be. Depending upon whether the graphics benchmark results skew marginally higher than I'm expecting, a higher clock of maybe 533 MHz (probably not higher than 567 MHz, though) might fit for the GPU with a 1.6 GHz clock (with an outside chance of 1.7 GHz) for the CPU.
Last edited by a moderator:
anexanhume
Veteran
How are they spending those other transistors though? Seems more cache and 128-bit memory interface most likely candidates.
I wonder if significantly more was added to the SRAM pool and/or to other on-chip memories to account for the extra transistors on this one.
edit: After editing my previous post, I missed the last response... But, yeah, extra transistors for on-chip memories and also the memory interface should account for some of the additional complexity. Still, though, there's a lot about the design and density of the SoC that doesn't seem to be well understood yet.
edit: After editing my previous post, I missed the last response... But, yeah, extra transistors for on-chip memories and also the memory interface should account for some of the additional complexity. Still, though, there's a lot about the design and density of the SoC that doesn't seem to be well understood yet.
Last edited by a moderator:
The memory interface is going to be an insignificant fraction of that 1B increase...